AI-Benchmark

Have some questions regarding the scores? Faced some issues? Want to discuss the results? Welcome to our new AI Benchmark Forum!

	Phones \| Mobile SoCs \| IoT \| Efficiency	Performance Ranking	Desktop GPUs and CPUs

Processor	AI Accelerator	Year	Lib	INT8 CNNs	INT8 Transformer	INT8 Accuracy	FP16 CNNs	FP16 Transformer	FP16 Accuracy	INT16 CNNs	INT8 Parallel	FP16 Parallel	AI Score

Dimensity 9600 Pro	NPU 1090	2026	mm	5957	6114	90.9	5290	2465	95.2	1183	1080	1601	22631
Dimensity 9600 M	NPU 990	2025	mm	5479	4313	90.7	4966	1061	83.8	941	861	1305	18306
Snapdragon 8 Elite Gen 5	Hexagon DSP / HTP Gen 5	2025	qhqh	5575	6310	85.7	3282	1626	96.6	307	1153	1025	17796
Snapdragon 8 Elite	Hexagon DSP / HTP Gen 4	2024	qhqh	4754	4995	85.7	2861	1367	96.6	229	772	664	14459
Snapdragon 8 Gen 3	Hexagon DSP / HTP Gen 3	2023	qhqh	3144	3629	85.7	1873	1030	96.5	155	629	557	10213
Snapdragon 8 Gen 2	Hexagon DSP / HTP Gen 2	2022	qhqh	2487	2560	85.7	1573	722	96.5	0.3	321	283	7361
Snapdragon 8+ Gen 1	Hexagon DSP / HTP	2022	qhqh	1683	2093	80.2	790	384	98.6	0	333	130	4898
Snapdragon 8 Gen 1	Hexagon DSP / HTP	2021	qhqh	1581	1815	77.4	750	373	98.6	0.3	163	103	4262
Dimensity 8400	NPU 880	2024	mm	1357	409	80.9	1259	83	86.5	332	205	337	3951
Dimensity 8300	NPU 780	2023	mm	1189	186	86.3	1010	39	89.9	0.1	232	253	2948
Snapdragon 888	DSP (Hex. 780) + GPU (Adreno 660)	2020	qhqg	1068	1150	77.4	201	52	70	0.3	137	26	2261
Dimensity 9200	APU 690	2022	mm	744	65	77.1	661	45	99.6	0	106	131	1722
Snapdragon 7 Gen 1	Hexagon DSP / HTP	2022	qhqh	484	580	75.5	260	137	98.6	0	74	41	1386
Dimensity 9000	APU 590	2021	mm	519	40	80.7	506	23	90.5	0	114	163	1339
Google Tensor G4	Google Tensor TPU 2.0	2024	nn	426	104	82.4	368	47	89.8	0.1	287	74	1301
Google Tensor G3	Google Tensor TPU 2.0	2023	nn	401	79	82.4	367	30	88.9	0	272	72	1207
Google Tensor G2	Google Tensor TPU 2.0	2022	nn	260	67	82.4	382	16	88.3	0	107	171	1003
Exynos 2200	GPU (Xclipse 920)	2022	gg	121	96	98.6	296	57	89.6	0	15	47	641
Kirin 9000 5G	GPU (Mali-G78 MP24)	2020	gg	107	116	98.6	263	56	89.6	0	14	49	616
Google Tensor	Google Tensor TPU	2021	nn	112	113	80.3	271	31	90.4	0	14	55	609
Snapdragon 865	DSP (Hex. 698) + GPU (Adreno 650)	2019	hg	91	98	77.4	215	52	88.2	0	9	35	476
Exynos 2100	GPU (Mali-G78 MP14)	2021	gg	79	93	98.6	176	46	89.7	0	13	50	464
Dimensity 1000+	GPU (Mali-G77 MC9)	2019	gg	59	110	98.6	140	39	89.7	0	11	38	402
Snapdragon 855	DSP (Hex. 690) + GPU (Adreno 640)	2018	hg	75	65	76.9	159	40	88.2	0	8.6	28	356
Kirin 980	GPU (Mali-G76 MP10)	2018	gg	58	69	98.6	144	39	89.6	0	5.9	20	343
Dimensity 1000+	APU 3.0 (6 cores)	2019	nn	67	22	87.2	132	7.8	83.3	0	8.9	44	269
Snapdragon 845	DSP (Hex. 685) + GPU (Adreno 630)	2018	hg	42	40	76.7	121	27	88.2	0	4.3	22	248
Dimensity 800	GPU (Mali-G57 MC4)	2020	gg	36	67	98.6	88	29	89.7	0	4.9	16	244
Snapdragon 835	DSP (Hex. 682) + GPU (Adreno 540)	2017	hg	36	23	78.1	91	13	88.2	0	4.7	15	176
Dimensity 800	APU 3.0 (4 cores)	2020	nn	55	22	85.7	70	2.8	83	0	6.8	16	164
Snapdragon 810	GPU (Adreno 430)	2014	gg	20	4.5	98.6	37	2.3	79.1	0	1.6	5	71
Snapdragon 625	GPU (Adreno 506)	2016	gg	4.2	9.6	98.6	13	9.3	88.2	0	0.7	2.2	41
Snapdragon 625	8x2.0 GHz Cortex-A53	2016	cc	6.8	12	100	7.1	12	100	0	1.8	1.5	41

xy - This device accelerates quantized [x] and floating-point [y] models using:

n - Android NNAPI

g - TensorFlow Lite GPU Delegate (OpenCL / OpenGL / Metal based)

h - Qualcomm Hexagon NN Direct Delegate

qh - Qualcomm QNN HTP Delegate

qd - Qualcomm QNN DSP Delegate

qg - Qualcomm QNN GPU Delegate

a - Arm NN Delegate

e - Samsung ENN Delegate

m - MediaTek Neuron Delegate

c - TensorFlow Lite / NNAPI default CPU backend

i - Apple CoreML Delegate

coreml - Apple CoreML Runtime

1 - This SoC might be using unofficial / prototype hardware or drivers

2 - These are the results of an early prototype. The results of the commercial SoC might be different

ETH Zurich, Switzerland